Xref #21255
I had a third idea that I had been meaning to write up but this could be a good plan in the meantime

Ah, right, I remembered a discussion but forgot it was a full issue.

Yeah, looks like this would do very little of what you want : ) It is a generic advertisement with none of the details.

I don't think an admonition would be the way to go for presenting details like the table in #21255 (comment); it is more of a way to get the word out that there's some sort of experimental support.

#21255 is not going to happen in this release so I agree this could be a good idea. Also we could probably do with updating https://scipy.github.io/devdocs/dev/api-dev/array_api.html#currently-supported-functionality

Also we could probably do with updating

I added at least a partial list of updates in the release notes.

Maybe we just include the admonitions at a module level and have it link to the list. Maybe we could add it to functions where other array types would offer big advantages, but it doesn't have to be added to every last function in stats, for instance.

Thanks for starting this @mdhaber. I think we need an admonition of some sort (either per module or per function), so this could be a start. The main thing is indeed how one would add details to the admonition (e.g., supported by CuPy, not by xxx). Providing a "test this please" request without those details in the 1.15.0 docs may get us a whole bunch of bug reports for things we already know don't work.

+1 for having an admonition that links to the complete (updated 😄 ) table in another page.

Also, the admonition as implemented here would allow for text to be defined in the body of the admonition to be appended to the default text. So something like

.. arrayapi:: function
	This function is supported by CuPy but not PyTorch.

would result in an admonition with title Array API Standard Support and body

This function has experimental support for Python Array API Standard
compatible backends other than NumPy. Please consider testing this
feature by setting an environment variable ``SCIPY_ARRAY_API=1`` and
providing PyTorch, JAX, or CuPy arrays as array arguments. See <link> for more information.
This function is supported by CuPy but not PyTorch.